Artificial Intelligence is transforming the healthcare industry and can lead to improved outcomes such as lower costs, but we need to address the fact that data can fail in ways that biological tissues cannot. Data is messy but it’s also fundamental to building AI engines for interpretation, such as progress overtime. We know that electronic records don’t tell the whole story. According to Pat Baird, Senior Regulatory Specialist at Philips, data brought into health ecosystems needs to be trustable, interoperable and linkable.
Here’s My Take: All data is bias, from the way it is captured to the way the data set is represented. The bias factors need to be transparent so that alternative interpretations of the results can be considered. The quality of the data has to be challenged: is it correct, complete or relevant? Too frequently, the priority is on getting the numbers, which results in skipping over the context of the capture and the source. Analytic tools applied to data sets require that the insights are grounded in context, and to be pertinent, the data needs to be interoperable. Algorithms in one patient population may not be applicable to another.
Reference: Trust and the Impact of AI on Health Care, CES 2021